Optimal design and management of local access networks

This dissertation addresses managerial and design issues associated with computer communication networks and uses mathematical modeling as an investigation tool. The primary focus is on optimal design and management of local access networks (LACN) and their interconnection to wide area networks (WAN).;The state of the art of network design and management is reviewed in detail. A subset of these current issues form the core of this dissertation. Research problems solved are practical extensions to the classical capacitated concentrator location problem (CCLP). Two major problems solved include the Generalized Capacitated Concentrator Location Problem (GCCLP), and the Robust Capacitated Concentrator Location Problem (RCCLP). GCCLP extends CCLP by simultaneously imposing both port and processing capacity restrictions on concentrators. RCCLP extends CCLP by incorporating data uncertainty in traffic demand imposed by LACN nodes and associated communication costs. This dissertation is a pioneering work for both GCCLP and RCCLP. All prior research work on CCLP are deterministic and have included only one type of capacity constraint, which by itself makes the problem NP-complete.;The motivation for the research work on GCCLP stems from the fact that in practice either type of capacity restrictions on concentrators could become binding in different situations. In this dissertation, two different solution algorithms are developed to solve GCCLP. In specific, a conventional Lagrangian relaxation based heuristic and a meta-search technique referred to as Greedy Randomized Adaptive Search Procedure (GRASP) are developed and their performance in solving GCCLP is compared.;In addressing RCCLP, a mathematical model and a non-deterministic robust optimization technique are developed. This facilitates a cost efficient network that is relatively insensitive (i.e., robust) to the potential changes in end-user demands and to the time varying communication costs. The presence of uncertainty in the input data adds a new dimension not yet addressed in the telecommunication literature. The robust approach requires less information than traditional stochastic methods and hence has a greater implementation appeal. A summary and discussion of future research issues conclude this dissertation.